A structured POE catalyst dataset designed for modeling rather than just storage.
AI Research Engineer 路 Computational Mathematics 路 Multilingual Builder
James Han
I turn mathematical thinking, machine learning workflows, and product engineering into intelligent systems that are clear enough to use and rigorous enough to trust.
Professional identity
James Han · Han Jiajun · 韩嘉骏
A single technical profile connecting AI research, computational mathematics, multilingual systems, and product engineering.
Profile
A technical profile with research depth and product instincts.
Carnegie Mellon graduate in Computational and Applied Mathematics, currently working on AI opportunities, catalyst datasets, and ML workflows at Shanghai Research Institute of Chemical Industry.
Impact Matrix
Numbers that make the work tangible
Feature retrieval and analysis supporting GMV-driving decision workflows.
A multilingual lens for measuring bias in multimodal NLP systems.
Iterative evaluation that improved companion personality quality.
Focus
Where the resume becomes a story
AI for real domains
Translate messy cross-team needs into datasets, targets, features, and deliverables that machine learning can actually act on.
Product-grade tools
Build maintainable web apps, admin platforms, and databases that make technical operations easier to query, manage, and export.
Research taste
Work across proof theory, fair multilingual NLP, and AI companion behavior with the patience to define the right evaluation.
Case Studies
Concrete work across modeling, platforms, and evaluation
A
Catalyst performance modeling
Built a POE catalyst dataset with standardized fields and ML workflows that predicted polymer performance from catalyst and reaction features.
R虏 0.97 路 12 fields 路 SAIB
Calibration data platform
Developed LukaLEO with Flask, SQLAlchemy, PostgreSQL, and Flask-Admin to centralize instrument calibration workflows.
Flask 路 PostgreSQL 路 CSV exportC
Multilingual AI fairness
Designed fairness evaluation for multimodal NLP systems across nine languages, turning qualitative bias concerns into measurable experiments.
9 languages 路 NLP 路 evaluationResearch Method
A repeatable way to build useful AI systems
Model the domain
Start with field definitions, experimental context, and the constraints that matter to the user.
Measure the signal
Use validation splits, fairness probes, and qualitative review to separate real progress from noise.
Ship the workflow
Turn models and datasets into interfaces, exports, dashboards, and repeatable operating procedures.
ML
Polymer prediction
NLP
Fairness across languages
Visual Lab
A portfolio that behaves more like an interface than a document
The page layers generated research imagery, WebGL particles, animated cards, scroll reveals, and concrete resume data into one cohesive technical narrative.
WebGL Field
3D Cards
AI Visuals
Scroll Motion
Experience
Industry work across AI, data, and web systems
AI Research Engineer 路 Shanghai Research Institute of Chemical Industry
Coordinated with 9 internal departments to identify AI opportunities, curated a structured POE catalyst dataset with 12 standardized fields, and built ML workflows predicting polymer performance from catalyst and reaction features.
- Converted business and research needs into executable AI tasks, data requirements, and deliverables.
- Achieved up to R虏 = 0.97 on held-out validation for polymer performance prediction.
Software Development Intern 路 Bio-Techne
Developed LukaLEO, a web application that centralizes access to calibration data and makes instrument information easier to manage accurately.
- Engineered PostgreSQL data models with Flask and SQLAlchemy for maintainable calibration records.
- Delivered a Flask-Admin platform for uploads, calibration views, and CSV exports.
Data Analyst Intern 路 Zero-One Fission Digital Technology
Used SQL and feature analysis across 1000+ product and user features, supporting a LightGBM-based decision tree workflow to identify the top 10 drivers for Watsons online market GMV.
Research
Proofs, fairness, and AI personality
Sep 2024 - Jul 2025
Lambda Calculus and Proof Theory
Created proof-driven notes on normalization and reduction strategies, studying typed and untyped lambda calculus, Church-Rosser results, and formal verification implications.Sep 2023 - Jan 2024
Fairness in Multimodal and Multilingual NLP
Designed fairness evaluation for text-image systems across 9 languages, translating qualitative bias concerns into measurable experiments.Apr 2023 - May 2024
Companion Personality Development for AI Systems
Preprocessed and augmented datasets with TensorFlow, improving persona-consistency pass rate from 72% to 86% and reducing repetitive responses by around 30%.Skills
A toolkit that moves from theorem to interface
Languages & Data
Tools & Systems
Languages
Contact